Sei sulla pagina 1di 42

When Do Stop-Loss Rules Stop Losses?

$
Kathryn M. Kaminski
1,
, Andrew W. Lo
2,
Abstract
We propose a simple analytical framework to measure the value added or subtracted by stop-
loss rulespredetermined policies that reduce a portfolios exposure after reaching a certain
threshold of cumulative losseson the expected return and volatility of an arbitrary port-
folio strategy. Using daily futures price data, we provide an empirical analysis of stop-loss
policies applied to a buy-and-hold strategy using index futures contracts. At longer sam-
pling frequencies, certain stop-loss policies can increase expected return while substantially
reducing volatility, consistent with their objectives in practical applications.
Keywords: Investments; Portfolio Management; Risk Management; Asset Allocation;
Performance Attribution; Behavioral Finance
JEL Classication: G11, G12
$
The views and opinions expressed in this article are those of the authors, and do not necessarily represent
the views and opinions of AlphaSimplex Group, MIT, or any of their aliates or employees. The authors
make no representations or warranty, either expressed or implied, as to the accuracy or completeness of the
information, nor are they recommending that this article serve as the basis for any investment decision
this article is for information purposes only. We thank Dimitris Bertsimas, Margret Bjarnadottir, Michael
Brennan, Tom Brennan, Florian Ederer, Mike Epstein, Jasmina Hasanhodzic, Dirk Jenter, Carola Frydman,
Leonid Kogan, Gustavo Manso, Stewart Myers, Bernhard Nietert, Jun Pan, Sebastian Pokutta, Michael
Stutzer, Svetlana Sussman, and participants at the EURO2006 Conference, the INFORMS 2006 Annual
Meeting, the MIT Sloan Finance Lunch, the MIT Sloan Finance Seminar, and SSgA for helpful comments and
discussion. Research support from the MIT Laboratory for Financial Engineering is gratefully acknowledged.

Please direct all correspondence to: Andrew W. Lo, MIT Sloan School of Management, 100 Main Street,
E62618, Cambridge, MA 02142.
1
Alpha K Capital LLC.
2
Harris & Harris Group Professor, MIT Sloan School of Management; director, Laboratory for Financial
Engineering; and Chief Investment Strategist, AlphaSimplex Group, LLC.
Preprint submitted to Journal of Financial Markets June 28, 2013
1. Introduction
Thanks to the overwhelming dominance of the mean-variance portfolio optimization
framework pioneered by Markowitz (1952), Tobin (1958), Sharpe (1964), and Lintner (1965),
much of the investments literatureboth in academia and in industryhas focused on con-
structing well-diversied static portfolios using low-cost index funds. With little use for
active trading or frequent rebalancing, this passive perspective comes from the recognition
that individual equity returns are dicult to forecast and trading is not costless. The ques-
tionable benets of day-trading are unlikely to outweigh the very real costs of changing
ones portfolio weights. It is, therefore, no surprise that a buy-and-hold philosophy has
permeated the mutual-fund industry and the nancial planning profession.
3
However, this passive approach to investing is often contradicted by human behavior,
especially during periods of market turmoil.
4
behavioral biases sometimes lead investors
astray, causing them to shift their portfolio weights in response to signicant swings in
market indexes, often selling at the low and buying at the high. On the other hand,
some of the most seasoned investment professionals routinely make use of systematic rules
for exiting and re-entering portfolio strategies based on cumulative losses, gains, and other
technical indicators.
In this paper, we investigate the ecacy of such behavior in the narrow context of stop-
3
This philosophy has changed slightly with the recent innovation of a slowly varying asset allocation that
changes according to ones age (e.g., a lifecycle fund).
4
For example, psychologists and behavioral economists have documented the following systematic biases
in the human decisionmaking process: overcondence (Fischo and Slovic, 1980; Barber and Odean, 2001;
Gervais and Odean, 2001), overreaction (DeBondt and Thaler, 1986), loss aversion (Kahneman and Tver-
sky, 1979; Shefrin and Statman, 1985; Kahneman and Tversky, 1992; Odean, 1998), herding (Huberman
and Regev, 2001), psychological accounting (Kahneman and Tversky, 1981), miscalibration of probabilities
(Lichtenstein, Fischo, and Phillips, 1982), hyperbolic discounting (Laibson, 1997), and regret (Bell, 1982a,b;
Clarke, Krase, and Statman, 1994).
2
loss rules (i.e., rules for exiting an investment after some threshold of loss is reached and
re-entered after some level of gains is achieved). We wish to identify the economic motivation
for stop-loss policies so as to distinguish between rational and behavioral explanations for
these rules. While certain market conditions may encourage irrational investor behavior
(e.g., large rapid market declines) stop-loss policies are suciently ubiquitous that their use
cannot always be irrational.
This raises the question we seek to answer in this paper: When do stop-loss rules stop
losses? In particular, because a stop-loss rule can be viewed as an overlay strategy for a
specic portfolio, we can derive the impact of that rule on the return characteristics of the
portfolio. The question of whether or not a stop-loss rule stops losses can then be answered
by comparing the expected return of the portfolio with and without the stop-loss rule. If the
expected return of the portfolio is higher with the stop-loss rule than without it, we conclude
that the stop-loss rule does, indeed, stop losses.
Using simple properties of conditional expectations, we are able to characterize the
marginal impact of stop-loss rules on any given portfolios expected return, which we dene
as the stopping premium. We show that the stopping premium is inextricably linked to
the stochastic process driving the underlying portfolios return. If the portfolio follows a
random walk (i.e., independently and identically distributed returns) the stopping premium
is always negative. This may explain why the academic and industry literature has looked
askance at stop-loss policies to date. If returns are unforecastable, stop-loss rules simply
force the portfolio out of higher-yielding assets on occasion, thereby lowering the overall
expected return without adding any benets. In such cases, stop-loss rules never stop losses.
3
However, for non-random-walk portfolios, we nd that stop-loss rules can stop losses. For
example, if portfolio returns are characterized by momentum or positive serial correlation,
we show that the stopping premium can be positive and is directly proportional to the
magnitude of return persistence. Not surprisingly, if conditioning on past cumulative returns
changes the conditional distribution of a portfolios return, it should be possible to nd a
stop-loss policy that yields a positive stopping premium. We provide specic guidelines for
nding such policies under several return specications: mean reversion, momentum, and
Markov regime-switching processes. In each case, we are able to derive explicit conditions
for stop-loss rules to stop losses.
Of course, focusing on expected returns does not account for risk in any way. It may
be the case that a stop-loss rule increases the expected return but also increases the risk
of the underlying portfolio, yielding ambiguous implications for the risk-adjusted return of
a portfolio with a stop-loss rule. To address this issue, we compare the variance of the
portfolio with and without the stop-loss rule and nd that, in cases where the stop-loss rule
involves switching to a lower-volatility asset when the stop-loss threshold is reached, the
unconditional variance of the portfolio return is reduced by the stop-loss rule. A decrease in
the variance coupled with the possibility of a positive stopping premium implies that, within
the traditional mean-variance framework, stop-loss rules may play an important role under
certain market conditions.
To illustrate the empirical relevance of our analysis, we apply a simple stop-loss rule to
a standard asset-allocation problem of stocks versus bonds using daily futures data from
January 1993 to November 2011. We nd that stop-loss rules exhibit positive stopping
4
premiums over longer sampling frequencies over larger range of threshold values. These
policies also provide substantial reduction in volatility creating larger Sharpe ratios as a
result. This is a remarkable feat for a buy-high/sell-low strategy. For example in one
calibration, using stop loss over monthly intervals in daily data can increase the return by
1.5% and decrease the volatility by 5% causing an increase in the Sharpe Ratio by as much
as 20%. These results suggest that stop-loss rules may exploit conditional momentum eects
following periods of losses in equities. These results suggest that the random walk model is a
particularly poor approximation to U.S. stock returns and may improperly value the use of
non-linear policies such as stop-loss rules. This is consistent with Lo and MacKinlay (1999)
and others using various methods to examine limitations of the random walk.
2. Literature Review
Before presenting our framework for examining the performance impact of stop-loss rules,
we provide a brief review of the relevant portfolio-choice literature, and illustrate some of its
limitations to underscore the need for a dierent approach.
The standard approach to portfolio choice is to solve an optimization problem in a multi-
period setting, for which the solution is contingent on two important assumptions: the
choice of objective function and the specication of the underlying stochastic process for
asset returns. The problem was rst posed by Samuelson (1969) in discrete time and Merton
(1969) in continuous time, and solved in both cases by stochastic dynamic programming.
As the asset-pricing literature has grown, this paradigm has been extended in a number of
important directions.
5
5
For a comprehensive summary of portfolio choice see Brandt (2004). Recent extensions include pre-
5
However, in practice, household investment behavior seems to be at odds with nance
theory. In particular, Ameriks and Zeldes (2004) observe that most observed variation in an
individuals portfolio is attributed to a small number of signicant decisions they make as
opposed to marginal adjustments over time. Moreover, other documented empirical charac-
teristics of investor behavior include non-participation (Calvet, Campbell, and Sodini 2006);
under-diversication (Calvet, Campbell, and Sodini 2006); limited monitoring frequency and
trading (Ameriks and Zeldes 2004); survival-based selling decisions or a ight to safety
(Agnew 2003); an absence of hedging strategies (Massa and Simonov, 2004); and concen-
tration in simple strategies through mutual-fund investments (Calvet, Campbell, and Sodini
2006). Variations in investment policies due to characteristics such as age, wealth, and
profession have been examined as well.
6
In fact, in contrast to the over-trading phenomenon documented by Odean (1999) and
Barber and Odean (2000), Agnew (2003) asserts that individual investors actually trade
infrequently. By examining asset-class ows, she nds that investors often shift out of
equities after extremely negative asset returns into xed-income products, and concludes
that in retirement accounts, investors are more prone to exhibit a ight to safety instead
of explicit return chasing. Given that one in three of the workers in the United States
participate in 401(k) programs, it is clear that this ight to safety could have a signicant
dictability and autocorrelation in asset returns (Kim and Omberg, 1996; Liu, 1999; Campbell and Viceria,
1999; Brennan and Xia, 2001; Xia, 2001; and Wachter, 2002), model uncertainty (Barberis, 2000), transac-
tion costs (Balduzzi and Lynch, 1999), stochastic opportunity sets (Brennan, Schwartz, and Lagnado, 1997;
Campbell, Chan, and Viceria, 2003; and Brandt, Goyal, Santa-Clara, and Stroud, 2005), and behavioral
nance (see the references in footnote 4).
6
For example, lack of age-dependence in allocation, lower wealth and lower education with greater non-
participation and under-diversication, and greater sophistication in higher wealth investors have all been
considered (see Ameriks and Zeldes, 2004).
6
impact on market prices as well as demand. Consistent with Agnews ight-to-safety in
the empirical application of stop-loss, we nd momentum in long-term bonds as a result of
sustained periods of loss in equities. This suggests conditional relationships between stocks
and bonds, an implication that is also conrmed by our empirical results.
7
Although stop-loss rules are widely used, the corresponding academic literature is rather
limited. The market microstructure literature contains a number of studies about limit
orders and optimal order selection algorithms (Easley and OHara, 1991; Biais, Hillion,
and Spatt, 1995; Chakravarty and Holden, 1995; Handa and Schwartz, 1996; Harris and
Hasbrouck, 1996; Seppi, 1997; and Lo, MacKinlay, and Zhang, 2002). Carr and Jarrow
(1990) investigate the properties of a particular trading strategy that employs stop-loss
orders, and Shefrin and Statman (1985) and Tschoegl (1988) consider behavioral patterns
that may explain the popularity of stop-loss rules. However, to date, there has been no
systematic analysis of the impact of a stop-loss rule on an existing investment policy, an
oversight that we remedy in this paper.
3. A Framework for Analyzing Stop-Loss Rules
In this section, we outline a framework for measuring the impact of stop-loss policies on
investment performance. In Section 3.1, we begin by specifying a simple stop-loss policy
and deriving some basic statistics for its eect on an existing portfolio strategy. We describe
several generalizations and qualications of our framework in Section 3.2, and then apply
our framework in Section 4 to various return-generating processes including the Random
7
Although excess performance in long-term bonds may seem puzzling, from a historical perspective, the
deregulation of long-term government xed-income products in the 1950s could provide motivation for the
existence of these eects.
7
Walk Hypothesis, momentum and mean-reversion models, and regime-switching models.
3.1. Assumptions and denitions
Consider any arbitrary portfolio strategy P with returns {r
t
} that satisfy the following
assumptions:
(A1) The returns {r
t
} for the portfolio strategy P are stationary with nite mean and
variance
2
.
(A2) The expected return of P is greater than the risk-free rate r
f
, and let r
f
denote the risk premium of P.
Our use of the term portfolio strategy in Assumption (A1) is meant to underscore the
possibility that P is a complex dynamic investment policy, not necessarily a static basket of
securities. Assumption (A2) simply rules out perverse cases where stop-loss rules add value
because the safe asset has a higher expected return than the original strategy itself.
Now suppose an investor seeks to impose a stop-loss policy on a portfolio strategy. This
typically involves tracking the cumulative return R
t
(J) of the portfolio over a window of J
periods, where:
8
R
t
(J)
J

j=1
r
tj+1
(1)
and when the cumulative return crosses some lower boundary, reducing the investment in
P by switching into cash or some other safer asset. This heuristic approach motivates the
8
For simplicity, we ignore compounding eects and dene cumulative returns by summing simple returns r
t
instead of multiplying (1+r
t
). For purposes of dening the trigger of our stop-loss policy, this approximation
does not have signicant impact. However, we do take compounding into account when simulating the
investment returns of a portfolio with and without a stop-loss policy.
8
following denition:
Denition 1. A simple stop-loss policy S(, , J) for a portfolio strategy P with returns
{r
t
} is a dynamic binary asset-allocation rule {s
t
} between P and a risk-free asset F with
return r
f
, where s
t
is the proportion of assets allocated to P, and:
s
t

_

_
0 if R
t1
(J) < and s
t1
= 1 (exit)
1 if r
t1
and s
t1
= 0 (re-enter)
1 if R
t1
(J) and s
t1
= 1 (stay in)
0 if r
t1
< and s
t1
= 0 (stay out)
(2)
for 0. Denote by r
st
the return of portfolio strategy S, which is the combinaton of portfolio
strategy P and the stop-loss policy S, hence:
r
st
s
t
r
t
+ (1 s
t
)r
f
. (3)
Denition 1 describes a 0/1 asset-allocation rule between P and the risk-free asset F, where
100% of the assets are withdrawn from P and invested in F as soon as the J-period cumula-
tive return R
t
1
(J) reaches some loss threshold at t
1
. The stop-loss rule stays in place until
some future date t
2
1 > t
1
when P realizes a return r
t
2
1
greater than , at which point
100% of the assets are transferred from F back to P at date t
2
. Therefore, the stop-loss pol-
icy S(, , J) is a function of three parameters: the loss threshold , the re-entry threshold
, and the cumulative-return window J. Of course, the performance of the stop-loss policy
also depends on the characteristics of Flower risk-free rates imply a more signicant drag
on performance during periods when the stop-loss policy is in eect.
Note that the specication of the loss and re-entry mechanisms are dierent; the exit
decision is a function of the cumulative return R
t1
(J), whereas the re-entry decision involves
9
only the one-period Return, r
t1
. This is intentional, and motivated by two behavioral
biases. The rst is loss aversion and the disposition eect, in which an individual becomes
less risk-averse when facing mounting losses. The second is the snake-bite eect, in which
an individual is more reluctant to re-enter a portfolio after experiencing losses from that
strategy. The simple stop-loss policy in Denition 1 is meant to address both of these
behavioral biases in a systematic fashion.
To gauge the impact of the stop-loss policy S on performance, we dene the following
metric:
Denition 2. The stopping premium

(S) of a stop-loss policy S is the expected return


dierence between the stop-loss policy S and the portfolio strategy P:

E[r
st
] E[r
t
] = p
o
_
r
f
E[r
t
|s
t
= 0]
_
, (4)
where p
o
Prob(s
t
= 0) (5)
and the stopping ratio is the ratio of the stopping premium to the probability of stopping
out:

p
o
= r
f
E[r
t
|s
t
= 0] . (6)
Note that the dierence of the expected returns of r
st
and r
t
reduces to the product of the
probability of a stop-loss p
o
and the conditional expectation of the dierence between r
f
and
r
t
, conditioned on being stopped out. The intuition for this expression is straightforward:
the only times r
st
and r
t
dier are during periods when the stop-loss policy has been trig-
10
gered. Therefore, the dierence in expected return should be given by the dierence in the
conditional expectation of the portfolio with and without the stop-loss policyconditioned
on being stopped outweighted by the probability of being stopped out.
The stopping premium (4) measures the expected-return dierence per unit time between
the stop-loss policy S and the portfolio strategy P, but this metric may yield misleading
comparisons between two stop-loss policies that have very dierent parameter values. For
example, for a given portfolio strategy P, suppose S
1
has a stopping premium of 1% and
S
2
has a stopping premium of 2%; this suggests that S
2
is superior to S
1
. But suppose the
parameters of S
2
implies that S
2
is active only 10% of the time (i.e., one month out of every
10 on average), whereas the parameters of S
1
implies that it is active 25% of the time. On
a total-return basis, S
1
is superior, even though it yields a lower expected-return dierence
per-unit-time. The stopping ratio

/p
o
given in (6) addresses this scale issue directly by
dividing the stopping premium by the probability p
o
. The reciprocal of p
o
is the expected
number of periods that s
t
=0 or the expected duration of the stop-loss period. Multiplying
the per-unit-time expected-return dierence

by this expected duration 1/p


o
then yields
the total expected-return dierence

/p
o
between r
f
and r
t
.
The probability p
o
of a stop-loss is of interest in its own right because more frequent
stop-loss events imply more trading and, consequently, more transactions costs. Although
we have not incorporated transactions costs explicitly into our analysis, this can be done
easily by imposing a return penalty in (3):
r
st
s
t
r
t
+ (1 s
t
)r
f
|s
t
s
t1
|, (7)
11
where >0 is the one-way transactions cost of a stop-loss event. For expositional simplicity,
we shall assume =0 for the remainder of this paper.
Using the metrics proposed in Denition 2, we now have a simple way to answer the
question posed in our title: stop-loss policies can be said to stop losses when the correspond-
ing stopping premium is positive. In other words, a stop-loss policy adds value if and only
if its implementation leads to an improvement in the overall expected return of a portfolio
strategy.
Of course, this simple interpretation of a stop-loss policys ecacy is based purely on
expected return, and ignores risk. Risk matters because it is conceivable that a stop-loss
policy with a positive stopping premium generates so much additional risk that the risk-
adjusted expected return is less attractive with the policy in place than without it. This
may seem unlikely because by construction, a stop-loss policy involves switching out of
P into a risk-free asset, implying that P spends more time in higher-risk assets than the
combination of P and S. However, it is important to acknowledge that P and S are dynamic
strategies and static measures of risk such as standard deviation are not sucient statistics
for the intertemporal risk/reward trade-os that characterize a dynamic rational expectations
equilibrium (e.g., Merton, 1973; Lucas, 1978). Nevertheless, it is still useful to gauge the
impact of a stop-loss policy on volatility of a portfolio strategy P, as only one of possibly
many risk characteristics of the combined strategy. To that end, we have:
12
Denition 3. Let the variance dierence

2 of a stopping strategy be given by:

2 Var[r
st
] Var[r
t
] (8)
= E
_
Var[r
st
|s
t
]

+ Var
_
E[r
st
|s
t
]

E
_
Var[r
t
|s
t
]

Var
_
E[r
t
|s
t
]

(9)
= p
o
Var[r
t
|s
t
= 0] +
p
o
(1 p
o
)
_
_
r
f
E[r
t
|s
t
= 0]
_
2

_
E[r
t
|s
t
= 0]
1 p
o
_
2
_
(10)
From an empirical perspective, standard deviations are often easier to interpret, hence we
also dene the quantity

_
Var[r
st
] .
Given that a stop-loss policy can aect both the mean and standard deviation of the
portfolio strategy P, we can also dene the dierence between the Sharpe ratios of P with
and without S:

SR

E[r
st
] r
f

r
f

. (11)
However, given the potentially misleading interpretations of the Sharpe ratio for dynamic
strategies such as P and S, we refrain from using this metric for evaluating the ecacy of
stop-loss policies.
9
3.2. Generalizations and Qualications
The basic framework outlined in Section 3.1 can be generalized in many ways. For
example, instead of switching out of P and into a completely risk-free asset, we can allow
F to be a lower-risk asset but with some non-negligible volatility. More generally, instead
9
See Sharpe (1994), Spurgin (2001), and Lo (2002) for details.
13
of focusing on binary asset-allocation policies, we can consider a continuous function ()
[0, 1] of cumulative returns that declines with losses and rises with gains. Also, instead of a
single safe asset, we might consider switching into multiple assets when losses are realized,
or incorporate the stop-loss policy directly into the portfolio strategy P itself so that the
original strategy is aected in some systematic way by cumulative losses and gains. Finally,
there is nothing to suggest that stop-loss policies must be applied at the portfolio levelsuch
rules can be implemented security-by-security or asset-class by asset-class.
Of course, with each generalization, the gains in exibility must be traded o against
the corresponding costs of complexity and analytic intractability. These trade-os can only
be decided on a case-by-case basis, and we leave it to the reader to make such trade-os
individually. Our more modest objective in this paper is to provide a complete solution for
the leading case of the simple stop-loss policy in Denition (1). From our analysis of this
simple case, a number of generalizations should follow naturally, some of which are explored
in Kaminski (2006).
However, an important qualication regarding our approach is the fact that we do not
derive the simple stop-loss policy from any optimization problemit is only a heuristic,
albeit a fairly popular one among many institutional and retail investors. This is a distinct
departure from much of the asset-pricing literature in which investment behavior is modelled
as the outcome of an optimizing individual seeking to maximize his expected lifetime utility
by investing in a nite set of securities subject to a budget constraint (e.g., Merton, 1971).
While such a formal approach is certainly preferable if the consumption/investment problem
is well posed. For example, if preferences are given and the investment opportunity set is
14
completely specied, the simple stop-loss policy can still be studied in the absence of such
structure.
Moreover, from a purely behavioral perspective, it is useful to consider the impact of a
stop-loss heuristic even if it is not derived from optimizing behavior, precisely because we
seek to understand the basis of such behavior. Of course, we can ask the more challenging
question of whether the stop-loss heuristic can be derived as the optimal portfolio rule for
a specic set of preferences, but such inverse-optimal problems become intractable very
quickly (e.g., Chang, 1988). Instead, we have a narrower set of objectives in this paper:
to investigate the basic properties of simple stop-loss heuristics without reference to any
optimization problem, and with as few restrictions as possible on the portfolio strategy P
to which the stop-loss policy is applied. The benets of our narrower focus are the explicit
analytical results described in Section 4, and the intuition that they provide for how stop-loss
mechanisms add or subtract value from an existing portfolio strategy.
Although this approach may be more limited in the insights it can provide to the invest-
ment process, the siren call of stop-loss rules seems so universal that we hope to derive some
useful implications for optimal consumption and portfolio rules from our analysis. Moreover,
the idea of overlaying one set of heuristics on top of an existing portfolio strategy has a cer-
tain operational appeal that many institutional investors have found so compelling recently
(e.g., so-called portable alpha strategies). Overlay products can be considered a general
class of superposition strategies, which is explored in more detail in Kaminski (2006).
15
4. Analytical Results
Having dened the basic framework in Section 3 for evaluating the performance of simple
stop-loss rules, we now apply them to several specic return-generating processes for {r
t
},
including the Random Walk Hypothesis in Section 4.1, mean-reversion and momentum pro-
cesses in Section 4.2, and a statistical regime-switching model in Section 4.3. The simplicity
of our stop-loss heuristic will allow us to derive explicit conditions under which stop-loss
policies can stop losses in each of these cases.
4.1. The Random Walk Hypothesis
Since the Random Walk Hypothesis is one of the most widely used return-generating
processes in the nance literature, any analysis of stop-loss policies must consider this leading
case rst. Given the framework proposed in Section 3, we are able to derive a surprisingly
strong conclusion about the ecacy of stop-loss rules:
Proposition 1. If {r
t
} satises the Random Walk Hypothesis so that:
r
t
= +
t
,
t
IID
White Noise(0,
2

), (12)
then the stop-loss policy has the following properties:

= p
o
(r
f
) = p
o
. (13a)

p
o
= . (13b)

2 = p
o

2
+ p
o
(1 p
o
)
2
. (13c)

SR
=

2 +
2
. (13d)
16
Proof: See Appendix A.1.
Proposition 1 shows that, for any portfolio strategy with an expected return greater
than the risk-free rate r
f
, the Random Walk Hypothesis implies that the stop-loss policy
will always reduce the portfolios expected return since

0. In the absence of any


predictability in {r
t
}, whether or not the stop-loss is activated has no informational content
for the portfolios returns; hence, the only eect of a stop-loss policy is to replace the portfolio
strategy P with the risk-free asset when the strategy is stopped out, thereby reducing the
expected return by the risk premium of the original portfolio strategy P. If the stop-loss
probability p
o
is large enough and the risk premium is small enough, (13) shows that the
stop-loss policy can also reduce the volatility of the portfolio.
The fact that there are no conditions under which the simple stop-loss policy can add
value to a portfolio with IID returns may explain why stop-loss rules have been given so little
attention in the academic nance literature. The fact that the Random Walk Hypothesis
was widely accepted in the 1960s and 1970s, and considered to be synonymous with market
eciency and rationality, eliminated the motivation for stop-loss rules altogether. In fact, our
simple stop-loss policy may be viewed as a more sophisticated version of the lter rule that
was tested extensively by Alexander (1961) and Fama and Blume (1966). Their conclusion
that such strategies did not produce any excess prots was typical of the outcomes of many
similar studies during this period.
However, despite the lack of interest in stop-loss rules in academic studies, investment
professionals have been using such rules for many years, and part of the reason for this di-
chotomy may be the fact that the theoretical motivation for the Random Walk Hypothesis is
17
stronger than the empirical reality. In particular, Lo and MacKinlay (1988) presented com-
pelling evidence against the Random Walk Hypothesis for weekly U.S. stock-index returns
from 1962 to 1985, which has subsequently been conrmed and extended to other markets
and countries by a number of other authors. In the next section, we shall see that, if asset-
returns do not follow random walks, there are several situations in which stop-loss policies
can add signicant value to an existing portfolio strategy.
4.2. Mean Reversion and Momentum
In the 1980s and 1990s, several authors documented important departures from the Ran-
dom Walk Hypothesis for U.S. equity returns (e.g., Fama and French,1988; Lo and MacKin-
lay, 1988, 1990, 1999; Poterba and Summers, 1988; Jegadeesh, 1990; Lo, 1991; and Jegadeesh
and Titman, 1993) and, in such cases, the implications for the stop-loss policy can be quite
dierent than in Proposition 1. To see how, consider the simplest case of a non-random-walk
return-generating process, the AR(1):
r
t
= + (r
t1
) +
t
,
t
IID
White Noise(0,
2

) , (1, 1) (14)
where the restriction that lies in the open interval (1, 1) is to ensure that r
t
is a stationary
process (see Hamilton, 1994).
This simple process captures a surprisingly broad range of behavior depending on the
single parameter , including the Random Walk Hypothesis ( = 0), mean reversion (
(1, 0)), and momentum ( = (0, 1)). However, the implications of this return-generating
process for our stop-loss rule are not trivial to derive because the conditional distribution of
r
t
, conditioned on R
t1
(J), is quite complex. For example, according to Denition (4), the
18
expression for the stopping premium

is given by:

= p
o
(r
f
E[r
t
|s
t
= 0]) (15)
but the conditional expectation E[r
t
|s
t
= 0] is not easy to evaluate in closed-form for an
AR(1). For =0, the conditional expectation is likely to dier from the unconditional mean
since past returns do contain information about the future, but the exact expression is
not easily computable. Fortunately, we are able to obtain a good rst-order approximation
under certain conditions, yielding the following result:
Proposition 2. If {r
t
} satises an AR(1) (14), then the stop-loss policy (2) has the follow-
ing properties:

p
o
= + + (, , J) (16)
and for > 0 and reasonable stop-loss parameters, it can be shown that (, , J) 0, which
yields the following lower bound:

p
o
+ , (17)
Proof: See Appendix A.2.
Proposition 2 shows that the impact of the stop-loss rule on expected returns is the sum
of three terms: the negative of the risk premium, a linear function of the autoregressive
parameter , and a remainder term. For a mean-reverting portfolio strategy, <0; hence,
the stop-loss policy hurts expected returns to a rst-order approximation. This is consistent
19
with the intuition that mean-reversion strategies benet from reversals, thus a stop-loss
policy that switches out of the portfolio after certain cumulative losses will miss the reversal
and lower the expected return of the portfolio. On the other hand, for a momentum strategy,
>0, in which case there is a possibility that the second term dominates the rst, yielding a
positive stopping premium. This is also consistent with the intuition that in the presence of
momentum, losses are likely to persist, therefore, switching to the risk-free asset after certain
cumulative losses can be more protable than staying fully invested.
In fact, (17) implies that a sucient condition for a stop-loss policy with reasonable
parameters to add value for a momentum-AR(1) return-generating process is:

SR, (18)
where SR is the usual Sharpe ratio of the portfolio strategy. In other words, if the return-
generating process exhibits enough momentum, then the stop-loss rule will indeed stop losses.
This may seem like a rather high hurdle, especially for hedge-fund strategies that have Sharpe
ratios in excess of 1.00. However, note that (18) assumes that the Sharpe ratio is calibrated
at the same sampling frequency as . Therefore, if we are using monthly returns in (14), the
Sharpe ratio in (18) must also be monthly. A portfolio strategy with an annual Sharpe ratio
of 1.00, annualized in the standard way by multiplying the monthly Sharpe ratio by

12,
implies a monthly Sharpe ratio of 0.29, which is still a signicant hurdle for but not quite
as imposing as 1.00.
10
10
Of course, the assumption that returns follow an AR(1) makes the usual annualization factor of

12
incorrect, which is why we use the phrase annualized in the standard way. See Lo (2002) for the proper
method of annualizing Sharpe ratios in the presence of serial correlation.
20
4.3. Regime-Switching Models
Statistical models of changes in regime, such as the Hamilton (1989) model, are parsimo-
nious ways to capture apparent nonstationarities in the data, such as sudden shifts in means
and variances. Although such models are, in fact, stationary, they do exhibit time-varying
conditional means and variances, conditioned on the particular state that prevails. Moreover,
by assuming that transitions from one state to another follow a time-homogenous Markov
process, regime-switching models exhibit rich time-series properties that are surprisingly dif-
cult to replicate with traditional linear processes. Regime-switching models are particularly
relevant for stop-loss policies because one of the most common reasons investors put forward
for using a stop-loss rule is to deal with a signicant change in market conditions, such as
October 1987 or August 1998. To the extent that this motivation is genuine and appropriate,
we should see signicant advantages to using stop-loss policies when the portfolio return {r
t
}
follows a regime-switching process.
More formally, let r
t
be given by the following stochastic process:
r
t
= I
t
r
1t
+ (1 I
t
)r
2t
, r
it
IID
N(
i
,
2
i
) , i = 1, 2 (19a)
A
_
_
_
_
I
t+1
=1 I
t+1
=0
I
t
=1 p
11
p
12
I
t
=0 p
21
p
22
_
_
_
_
(19b)
where I
t
is an indicator function that takes on the value 1 when state 1 prevails and 0
when state 2 prevails, and A is the Markov transition probabilities matrix that governs the
transitions between the two states. The parameters of (19) are the means and variances of
the two states, (
1
,
2
,
2
1
,
2
2
), and the transition probabilities (p
11
, p
22
). Without any loss
21
in generality, we adopt the convention that state 1 is the higher-mean state so that
1
>
2
.
Given assumption (A2), this convention implies that
1
> r
f
, which is an inequality we will
use below. The six parameters of (19) may be estimated numerically via maximum likelihood
(Hamilton, 1994).
Despite the many studies in the economics and nance literatures that have implemented
the regime-switching model (19), the implications of regime-switching returns for the invest-
ment process has only recently been considered (see Ang and Bekaert, 2004). This is due,
in part, to the analytical intractability of (19)while the specication may seem simple, it
poses signicant challenges for even the simplest portfolio optimization process. However,
numerical results can easily be obtained via Monte Carlo simulation, and we provide such
results in Sections 5.
In this section, we investigate the performance of our simple stop-loss policy for this
return-generating process. Because of the relatively simple time-series structure of returns
within each regime, we are able to characterize the stopping premium explicitly:
Proposition 3. If {r
t
} satises the two-state Markov regime-switching process (19), then
the stop-loss policy (2) has the following properties:

= p
o,1
(r
f

1
) + p
o,2
(r
f

2
) (20)

p
o
= (1 p
o,2
)(r
f

1
) + p
o,2
(r
f

2
) (21)
22
where:
p
o,1
Prob ( s
t
=0, I
t
=1 ). (22a)
p
o,2
Prob ( s
t
=0, I
t
=0 ). (22b)
p
o,2

p
o,2
p
o
= Prob (I
t
=0 | s
t
=0) . (22c)
If the risk-free rate r
f
follows the same two-state Markov regime-switching process (19), with
expected returns r
f1
and r
f2
in states 1 and 2, respectively, then the stop-loss policy has the
following properties:

= p
o,1
(r
f1

1
) + p
o,2
(r
f2

2
). (23)

p
o
= (1 p
o,2
)(r
f1

1
) + p
o,2
(r
f2

2
) . (24)
The conditional probability p
o,2
can be interpreted as the accuracy of the stop-loss policy in
anticipating the low-mean regime. The higher is this probability, the more likely it is that
the stop-loss policy triggers during low-mean regimes (regime 2), which should add value to
the expected return of the portfolio as long as the risk-free asset-return r
f
is suciently high
relative to the low-mean expected return
2
.
In particular, we can use our expression for the stopping ratio

/p
o
to provide a bound
on the level of accuracy required to have a non-negative stopping premium. Consider rst the
case where the risk-free asset r
f
is the same across both regimes. For levels of p
o,2
satisfying
the inequality:
p
o,2


1
r
f

1

2
(25)
23
the corresponding stopping premium

will be non-negative. By convention,


1
>
2
, and
by assumption (A2),
1
>r
f
, therefore the sign of the right side of (25) is positive. If r
f
is less
than
2
, then the right side of (25) is greater than 1, and no value of p
o,2
can satisfy (25).
If the expected return of equities in both regimes dominates the risk-free asset, then the
simple stop-loss policy will always decrease the portfolios expected return, regardless of how
accurate it is. To see why, recall that returns are independently and identically distributed
within each regime, and we know from Section 4.1 that our stop-loss policy never adds value
under the Random Walk Hypothesis. Therefore, the only source of potential value-added
for the stop-loss policy under a regime-switching process is if the equity investment in the
low-mean regime has a lower expected return than the risk-free rate (i.e.,
2
<r
f
). In this
case, the right side of (25) is positive and less than 1, implying that suciently accurate
stop-loss policies will yield positive stopping premia.
Note that the threshold for positive stopping premia in (25) is decreasing in the spread

2
. As the dierence between expected equity returns in the high-mean and low-mean
states widens, less accuracy is needed to ensure that the stop-loss policy adds value. This
may be an important psychological justication for the ubiquity of stop-loss rules in practice.
If an investor possesses a particularly pessimistic view of the low-mean state, implying a large
spread between
1
and
2
, then our simple stop-loss policy may appeal to him even if its
accuracy is not very high.
5. Empirical Analysis
To illustrate the potential relevance of our framework for analyzing stop-loss rules, we
consider the performance of the simple stop-loss rule when applied to equity portfolios.
24
Given that most nancial hedging is done in the futures markets, we apply stop-loss rules
on equities using daily futures prices from January 5, 1993 until November 7, 2011. Similar
to how a futures position would be held, the futures prices represent a weighted basket of
prices over various maturities shorter in the curve, which are rolled over to avoid jumps in
prices near maturity. The IMM S&P futures contract is used for a position in U.S. Equities
and the 10-year CBT Treasury note futures contract. In this given sample period, the two
portfolios have a negative correlation of -17.29%. In Table 1, the basic statistical properties
of the two return series is detailed. In Table 2, the parameter estimates for a two-state
regime-switching model are also detailed.
Despite the net positive serial autocorrelation in the IMM S&P contract, over smaller
time intervals the serial autocorrelation seems to be time varying. In Figures 1 and 2, rolling
point estimates of serial autocorrelation are plotted using both a 150-day and a 75-day
window. These graphs suggests that there are periods when stocks can be either momentum
driven or mean reverting. Given the theoretical analysis in this paper, during periods of
sucient momentum stop-loss policies might provide a stopping premium. In the same
vein, during periods of mean reversion stop-loss policies may produce negative stopping
premiums. When a simple regime-switching model is applied to the IMM S&P contract
basket, the estimates also suggest that there are two regimes: one positive low volatility
regime and one negative higher volatility regime, which occurs less often. Given the analytic
results in Section 4, these parameter estimates indicate stopping rules, which can accurately
determine low performance regimes in stocks may add stopping premium.
25
5.1. Basic Results
To examine the performance of stop-loss rules, the approach can be applied from short-
term to longer time where the strategy can be applied daily (1 day), weekly (5 days), monthly
(20 days), and quarterly (60 days). For each frequency, the stopping windows will be mul-
tiples of 3, 5, and 10 times the length of the data frequency (daily, weekly, monthly, and
quarterly). The dierent strategy combinations include daily (3,1), (5,1), (10,1), weekly
(15,5), (25,5), (50,5), monthly (60,20), (100,20), (200,20), and quarterly (180,60), (300,60),
(600,60). Consistent with our theoretical framework, (i,j) represent the size of the stop-
ping window and the re-entry window is one period for each time frequency. The stopping
thresholds, (), will vary from -1.5 to -0.5 standard deviations from the mean at the relevant
frequency. For example, if the stopping window is three months long the stops will be set
relative to deviations from -1.5 to -0.5 standard deviations. To avoid data selection bias,
we review a large range of stops to demonstrate how the performance depends on threshold
choices. The re-entry threshold, (), will also be modulated to the data frequency and will
simply vary between -0.5 to 1 standard deviation from the mean. This approach is used to
allow for comparison across dierent frequencies of time. For example, a one standard devia-
tion stop-loss in weekly versus a one standard deviation stop-loss quarterly can be compared
to see how the time frequency impacts the results.
Given the large set of parameters we analyze in this experiment, it is not surprising
that the performance of these strategies varies. There are a few key trends in the results.
First, shorter term, lower frequency stop-loss policies have negative stopping premiums over
large ranges of parameters. Longer term stop-loss at frequencies above one month perform
26
better and can achieve positive stopping premiums. In Figure 3, the stopping premium
for all frequencies and combinations of threshold parameters is plotted as a function of the
stopping threshold with delta the re-entry threshold at a 0 standard deviation threshold.
In Figures 4 and 5, the empirical results for the change in Sharpe Ratio and change in
standard deviation demonstrate how for longer term stop-loss strategies the Sharpe ratio
can improve as standard deviation decreases. Second, the decision to exit and the stopping
threshold seems to have a larger impact on the variation in results than the re-entry threshold.
Putting these results together, the empirical results suggest that the use of longer term stop-
loss strategies might have improved performance consistent with anecdotal discussion of the
strategy in practice.
6. Conclusion
In this paper, we provide a concrete answer to the question of when do stop-loss rules stop
losses? The answer depends, of course, on the return-generating process of the underlying
investment for which the stop-loss policy is implemented, as well as the particular dynamics
of the stop-loss policy itself. If stopping losses is interpreted as having a higher expected
return with the stop-loss policy than without it, then for a specic binary stop-loss policy,
we derive various conditions under which the expected-return dierence, which we call the
stopping premium, is positive. We show that under the most common return-generating
process, the Random Walk Hypothesis, the stopping premium is always negative. The
widespread cultural anity for the Random Walk Hypothesis, despite empirical evidence
to the contrary, may explain the general indierence to stop-loss policies in the academic
nance literature.
27
However, under more empirically plausible return-generating processes such as momen-
tum or regime-switching models, we show that stop-loss policies can generate positive stop-
ping premia. When applied to a standard buy-and-hold strategy using daily U.S. futures
contracts from January 1993 to November 2011 where the stop-loss asset is U.S. long-term
bonds futures, we nd that at longer sampling frequencies, certain stop-loss policies add
value over a buy-and-hold portfolio while substantially reducing risk by reducing strategy
volatility, consistent with their objectives in practical applications. These empirical results
suggest important nonlinearities in aggregate stock and bond returns that have not been
fully explored in the empirical nance literature.
Our analytical and empirical results contain several points of intersection with the be-
havioral nance literature. First, the ight-to-safety phenomena, which is best illustrated
by events surrounding the default of Russian government debt in August 1998, may create
momentum in equity returns and increase demand for long-term bonds, creating positive
stopping premia as a result. Second, systematic stop-loss policies may prot from the dis-
position eect and loss aversion, the tendency to sell winners too soon and hold on to losers
too long. Third, if investors are ambiguity-averse, large negative returns may cause them to
view equities as more ambiguous which, in relative terms, will make long-term bonds seem
less ambiguous. This may cause investors to switch to bonds to avoid uncertainty about
asset returns.
More generally, there is now substantial evidence from the cognitive sciences literature
that losses and gains are processed by dierent components of the brain. These dierent
components provide a partial explanation for some of the asymmetries observed in exper-
28
imental and actual markets. In particular, in the event of a signicant drop in aggregate
stock prices, investors who are generally passive will become motivated to trade because
mounting losses will cause them to pay attention when they ordinarily would not. This
inux of uninformed traders, who have less market experience and are more likely to make
irrational trading decisions, can have a signicant impact on equilibrium prices and their
dynamics. Therefore, even if markets are usually ecient, on occasions where a signicant
number of investors experience losses simultaneously, markets may be dominated temporarily
by irrational forces. The mechanism for this coordinated irrationality is cumulative loss.
Of course, our ndings shed little light on the controversy between market eciency and
behavioral nance. The success of our simple stop-loss policy may be due to certain nonlinear
aspects of stock and bond returns from which our strategy happens to benet (e.g., avoiding
momentum on the downside and exploiting asymmetries in asset returns following periods
of negative cumulative returns). And from the behavioral perspective, our stop-loss policy is
just one mechanism for avoiding or anticipating the usual pitfalls of human judgment (e.g.,
the disposition eect, loss aversion, ambiguity aversion, and ight-to-safety).
In summary, both behavioral nance and rational asset-pricing models may be used to
motivate the apparent eectiveness of stop-loss policies, in addition to the widespread use
of such policies in practice. This underscores the importance of learning how to deal with
loss as an investor, of which a stop-loss rule is only one dimension. As dicult as it may be
to accept, for the many investors who lamented, after the subprime mortgage meltdown of
20072008, that if only I had gotten out sooner, I wouldnt have lost so much, they may
have been correct.
29
0 1000 2000 3000 4000 5000
0.5
0.4
0.3
0.2
0.1
0
0.1
0.2
0.3
75 day Rolling Serial Autocorrelation Estimates
time
A
u
t
o
c
o
r
r
e
l
a
t
i
o
n
Fig. 1. Rolling serial autocorrelation coecient estimates for the IMM S&P contact with a 75-day
Window
30
0 1000 2000 3000 4000 5000
0.4
0.3
0.2
0.1
0
0.1
0.2
0.3
150 day Rolling Serial Autocorrelation Estimates
A
u
t
o
c
o
r
r
e
l
a
t
i
o
n
time
Fig. 2. Rolling serial autocorrelation coecient estimates for the IMM S&P contact with a 150-day
Window
31
-0.04
-0.02
0
0.02
0.04
3,1 5,1 10,1 15,5 25,5 50,5 60,20 100,20 200,20 180,60 300,60 600,60 360,120 600,120 1200, 120

-0.1
-0.08
-0.06
-1.5 -1.2 -1.0 -0.8 -0.5
Fig. 3. Stopping premium, (

), from short-term (3,1) stop-loss strategies to longer-term


(1200,120) stop-loss strategies with exit thresholds, (), of 1.5, 1.2, 1, 0.8, and 0.5 standard devia-
tions from the mean with a 0 standard deviation re-entry threshold, ()
32
-0.1
0
0.1
0.2
0.3
0.4
3,1 5,1 10,1 15,5 25,5 50,5 60,20 100,20 200,20 180,60 300,60 600,60 360,120 600,120 1200, 120
Sh
-0.5
-0.4
-0.3
-0.2
-1.5 -1.2 -1 -0.8 -0.5
Fig. 4. Change in Sharpe Ratio, (
SR
), from short-term (3,1) stop-loss strategies to longer-term
(1200,120) stop-loss strategies with exit thresholds, (), of 1.5, 1.2, 1, 0.8, and 0.5 standard devia-
tions from the mean with a 0 standard deviation re-entry threshold, ()
33
-0.04
-0.03
-0.02
-0.01
0
3,1 5,1 10,1 15,5 25,5 50,5 60,20 100,20 200,20 180,60 300,60 600,60 360,120 600,120 1200,
120

-1.5000
-1.2000
-1.0000
-0.8000
-0.5000
-0.07
-0.06
-0.05
-0.04
-0.5000
Fig. 5. Change in standard deviation, (

), from short-term (3,1) stop-loss strategies to longer-


term (1200,120) stop-loss strategies with exit thresholds, (), set at 1.5, 1.2, 1, 0.8, and 0.5 standard
deviations from the mean with a 0 standard deviation re-entry threshold, ()
34
Ann. Ann. Skew Kurt Min Max Ann. MDD
Mean (%) SD (%) (%) (%) Sharpe (%)
IMM S&P 7.295 19.499 -0.06 0.12 14.24 -9.88 14.12 0.37 26.62
CBT 10YR TN 1.181 6.331 0.02 -0.09 6.11 -2.43 3.61 0.19 5.25
Table 1
Summary statistics for U.S. Equities (IMM S&P) and long-term U.S. Government Bonds (CBT 10
YR) from January 5, 1993 to November 7,2011. Statistics are annualized assuming 250 days per
year.
35

0

1

0

1
p
IMM S&P 22.3% -39.0% 10.3% 33.0% 0.72
Table 2
Annualized parameter estimates for a two-state Hamilton model estimated via the EM Algorithm
on the IMM S&P Futures Contracts from January 5, 1993 to November 7, 2011.
36
Appendix
In this Appendix, we provide proofs of Propositions 1 and 2 in Sections A.1 and A.2.
A.1. Proof of Proposition 1
The conclusion follows almost immediately from the observation that the conditional ex-
pectations in (4) and (6) are equal to the unconditional expectations because of the Random
Walk Hypothesis (conditioning on past returns provides no incremental information), hence:

= p
o
0 (.1)

p
o
= 0 (.2)
and the other relations follow in a similar manner.
A.2. Proof of Proposition 2
Let r
t
be a stationary AR(1) process:
r
t
= + (r
t1
) +
t
,
t
IID
White Noise(0,
2

), (1, 1) (.3)
We seek the conditional expectation of r
t
given that the process is stopped out. If we let
J be suciently large and = , s
t
=0 is equivalent to R
t1
(J) < and s
t1
=1 with
R
t2
(J). Using log returns, we have:
E[r
t
|s
t
= 0] = E[r
t
|R
t1
(J)<, R
t2
(J)] (.4)
= (1 ) + E[r
t1
+
t
|R
t1
(J)<, R
t2
(J)] (.5)
= (1 ) + E[r
t1
|R
t1
(J)<, R
t2
(J)]. (.6)
37
By denition R
t1
(J) r
t1
+ + r
tJ
and R
t2
(J) = r
t2
+ + r
tJ1
. Setting
y r
t2
+ + r
tJ
then yields:
E[r
t
|s
t
= 0] = (1 ) + E[r
t1
|R
t1
(J)<, R
t2
(J)] (.7)
= (1 ) + E
y
_
E[r
t1
|r
t1
< y, r
tJ1
y]

. (.8)
For J large enough, the dependence between r
tJ1
and r
t1
is of order o(
J
) 0, hence:
E
y
_
E[r
t1
|r
t1
< y]

E
r
tJ1
_
E[r
t1
|r
t1
<r
tJ1
]

(.9)
, (.10)
which implies:
E[r
t
|s
t
= 0] (1 ) + ( ). (.11)
. (.12)
38
References
Agnew, J., 2003. An analysis of how individuals react to market returns in one 401(K)
plan. Working paper, The College of William and Mary.
Alexander, S., 1961. Price movements in speculative markets: Trends or random walks.
Industrial Management Review 2, 726.
Ameriks, J., Zeldes, S., 2004. How do household portfolio shares vary with age?. Working
paper, Columbia University.
Ang, A., Bekaert, G., 2004. How regimes aect asset allocation. Financial Analysts Journal
60, 8699.
Balduzzi, P., Lynch, A., 1999. Transaction costs and predictability: Some utility cost
calculations. Journal of Financial Economics 52, 4778.
Barber, B., Odean, T., 2000. Trading is hazardous to your wealth: The common stock
investment performance of individual investors. Journal of Finance 55, 773806.
Barber, B., Odean, T., 2001. Boys will be boys: Gender, overcondence, and common
stock investment. Quarterly Journal of Economics 116, 261292.
Barberis, N., 2000. Investing in the long run when returns are predictable. Journal of
Finance 55, 225264.
Bell, D., 1982a. Regret in decision making under uncertainty. Operations Research 30,
961981.
Bell, D., 1982b. Risk premiums for decision regret. Management Science 29, 11561166.
Biais, B., Hillion, P., Spatt, C., 1995. An empirical analysis of the limit order book and
the order ow in the Paris Bourse. Journal of Finance 50, 16551689.
Brandt, M., 2010. Portfolio Choice Problems. In: Ait-Sahalia, Y., Hansen, L. (Eds.),
Handbook of Financial Econometrics, Volume 1: Tools and Techniques. Elsevier,
North Holland, pp. 269336.
Brandt, M., Goyal, A., Santa-Clara, P., Stroud, J., 2005. A Simulation approach to dynamic
portfolio choice with an application to learning about return predictability. Review of
Financial Studies 18, 831873.
Brennan, M., Xia, Y., 2001. Persistence, predictability, and portfolio planning. Working
paper.
Brennan, M., Schwartz, E., Lagnado, R., 1997. Strategic Asset Allocation. Journal of
Economic Dynamics and Control 21, 13771403.
Calvet, L., Campbell, J., Sodini, P., 2006. Down or out: Assesing the welfare costs of
household investment mistakes. Working Paper No. 12030, NBER.
Campbell, J., Chan, Y., Viceira, L., 2003. A multivariate model of strategic asset allocation.
Journal of Financial Economics 67, 4180.
39
Campbell, J., Lo, A., MacKinlay, C., 1997. The Econometrics of Financial Markets. Prince-
ton University Press, Princeton.
Campbell, J., Viceira, L., 1999. Consumption and portfolio decisions when expected returns
are time-varying. Quarterly Journal of Economics 114, 433495.
Carr, P., Jarrow, R., 1990. The stop-loss start-gain paradox and option valuation: A new
decomposition into intrinsic and time value. Review of Financial Studies 3, 469492.
Chakravarty, S., Holden, C., 1995. An integrated model of market and limit orders. Journal
of Financial Intermediation 4, 213241.
Chang, F., 1988. The inverse optimal problem: A dynamic programming approach. Econo-
metrica 56, 147172.
Clarke, R., Krase, S., Statman, M., 1994. Tracking errors, regret, and tactical asset alloca-
tion. Journal of Portfolio Management 20, 1624.
DeBondt, W., Thaler, R., 1986. Does the stock market overreact?. Journal of Finance 40,
793807.
Easley, D., OHara, M., 1991. Order form and information in securities markets. Journal
of Finance 46, 905927.
Fama, E., Blume, M., 1966. Filter rules and stock market trading prots. Journal of
Business 39, 226241.
Fama, E., French, K., 1988. Permanent and temporary components of stock prices. Journal
of Political Economy 96, 246273.
Fischo, B., Slovic, P., 1980. A little learning...: Condence in multicue judgment tasks.
In: Nickerson, R. (Eds.), Attention and Performance, VIII. Erlbaum, Hillsdale, pp.
779800.
Gervais, S., Odean, T., 2001. Learning to be overcondent. Review of Financial Studies
14, 127.
Hamilton, J., 1989. A new approach to the economic analysis of nonstationary time series
and the business cycle. Econometrica 57, 357384.
Hamilton, J., 1994. Time Series Analysis. Princeton University Press, Princeton.
Handa, P., Schwartz, R., 1996. Limit order trading. Journal of Finance 51, 18351861.
Harris, L., Hasbrouck, J.,1996. Market vs. limit-orders: The SuperDot evidence on order
submission strategy. Journal of Financial and Quantitative Analysis 31, 213231.
Huberman, G., Regev, T., 2001. Contagious speculation and a cure for cancer: A nonevent
that made stock prices soar. Journal of Finance 56, 387396.
Jegadeesh, N., 1990. Evidence of predictable behavior of security returns. Journal of
Finance 45, 881898.
Jegadeesh, N., Titman, S., 1993. Returns to buying winners and selling losers: Implications
for stock market eciency. Journal of Finance 48, 6591.
40
Kahneman, D., Tversky, A., 1979. Prospect theory: An analysis of decision under risk.
Econometrica 47, 263291.
Kahneman, D., Tversky, A., 1981. The framing of decisions and the psychology of choice.
Science 211, 453458.
Kahneman, D., Tversky, A., 1992. Advances in prospect theory: Cumulative representation
of uncertainty. Journal of Risk and Uncertainty 5, 297323.
Kaminski, K., 2006. General Superposition Strategies. unpublished manuscript.
Kim, T., Omberg, E., 1996. Dynamic non-myopic portfolio behavior. Review of Financial
Studies 9, 141161.
Laibson, D., 1997. Golden eggs and hyperbolic discounting. Quarterly Journal of Eco-
nomics 62, 443477.
Lichtenstein, S., Fischo, B., Phillips, L., 1982. Calibration of Probabilities: The State
of the Art to 1980. In: Kahneman, D., Slovic, P., Tversky, A. (Eds.), Judgment
Under Uncertainty: Heuristics and Biases. Cambridge University Press, New York,
pp. 306334.
Lintner, J., 1965. The valuation of risk assets and the selection of risky investments in
stock portfolios and capital budgets. Review of Economics and Statistics 47, 1337.
Liu, J., 1999. Portfolio selection in stochastic environments. unpublished working paper,
University of California, Los Angeles.
Lo, A., 1991. Long-term memory in stock market prices. Econometrica 59, 12791313.
Lo, A., 2002. The statistics of Sharpe ratios. Financial Analysts Journal 58, 3650.
Lo, A. and C. MacKinlay, 1988. Stock market prices do not follow random walks: Evidence
from a simple specication test. Review of Financial Studies 1, 4166.
Lo, A., MacKinlay, C., 1990. An econometric analysis of nonsynchronous trading. Journal
of Econometrics 45, 181212.
Lo, A., MacKinlay, C., 1999. A Non-Random Walk Down Wall Street. Princeton University
Press, Princeton.
Lo, A., MacKinlay, C., Zhang, J., 2002. Econometric models of limit-order executions.
Journal of Financial Economics 65, 3171.
Lucas, R., 1978. Asset pricing in an exchange economy. Econometrica 46, 14291446.
Markowitz, H., 1952. Portfolio selection. Journal of Finance 7, 7791.
Massa, M., Simonov, A., 2004. Hedging, familiarity and portfolio choice. CEPR Discussion
Paper No. 4789.
Merton, R. 1969. Lifetime portfolio selection under uncertainty; The continuous-time case.
Review of Economics & Statistics 51, 247257.
Merton, R., 1971. Optimum consumption and portfolio rules in a continuous-time model.
Journal of Economic Theory 3, 373413.
41
Merton, R., 1973. An intertemporal capital asset pricing model. Econometrica 41, 867887.
Odean, T., 1998. Are investors reluctant to realize their losses?. Journal of Finance 53,
17751798.
Odean, T., 1999. Do investors trade too much. American Economic Review 89, 12791298.
Poterba, J., Summers, L., 1988. Mean reversion in stock returns: Evidence and implica-
tions. Journal of Financial Economics 22, 2760.
Samuelson, P. 1969. Lifetime portfolio selection by dynamic stochastic programming. Re-
view of Economics and Statistics 51, 239246.
Seppi, D., 1997. Liquidity provision with limit orders and a strategic specialist. Review of
Financial Studies 10, 103150.
Sharpe, W., 1964. Capital asset prices: A theory of market equilibrium under conditions
of risk. Journal of Finance 19, 425442.
Sharpe, W., 1994. The Sharpe ratio. Journal of Portfolio Management Fall, 4958.
Shefrin, M., Statman, M., 1985. The disposition to sell winners too early and ride losers
too long: Theory and evidence. Journal of Finance 40, 777790.
Spurgin, R., 2001. How to game your Sharpe ratio. The Journal of Alternative Investments
4, 3846.
Tobin, J., 1958. Liquidity preference as behavior towards risk. Review of Economic Studies
25, 6586.
Tschoegl, A., 1988. The source and consequences of stop orders: A conjecture. Managerial
and Decision Economics 9, 8385.
Wachter, J., 2002. Portfolio and consumption decisions under mean-reverting returns: An
exact solution for complete markets. Journal of Financial and Quantitative Analysis
37, 6391.
Xia, Y., 2001. Learning about predictability: The eects of parameter uncertainty on
dynamic asset allocation. Journal of Finance 56, 205246.
42